Yao Hu

Xiaohongshu (Little Red Book) (China)

Unclaimed · AI-structured profileOpenAlex

Xiaohongshu researcher spanning RL for LLM post-training and LLM search agents; first visible works in 2026

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

No output score is computed: beyond 4 works indexed in 2026, the OpenAlex snapshot lacks a stable publication count for this identity. This axis would measure total and per-year publication volume once author attribution is confirmed.

Impact—

Impact cannot be scored: the snapshot reports h-index 0 and i10-index 0, consistent with a newly indexed profile rather than measured citation influence. This axis would aggregate citations per work, citing works, and venue visibility as evidence accumulates.

Focus0/100

The recorded focus score is 0, mirroring an evenly split profile: fieldDist allocates 0.5 to machine learning (reinforcement learning for LLM post-training) and 0.5 to natural language processing / LLM agents. OpenAlex topics reinforce the spread — Information Retrieval and Search Behavior (3), Topic Modeling (2), Recommender Systems and Techniques (2), Web Data Mining and Analysis (2), and Natural Language Processing Techniques (1) — five topic areas rather than one concentrated line.

Collaboration54/100

The snapshot records 14 distinct coauthors and a recurring core of 12 frequent collaborators — including Zhaokai Luo, Mu Chuan, Yong Zhang, and Dongyan Zhao — across the reviewed 2026 papers. The computed collaboration score is 54, consistent with mid-sized, team-based industry research. Both reviewed papers place Yao Hu in a middle author position rather than first or corresponding, so leadership signals are not yet visible in the catalog.

Activity47/100

Activity is recent and tightly concentrated: countsByYear records 4 works in 2026 and none in earlier years, and recent3yWorks equals 4, yielding an activity score of 47. With the entire visible record sitting in a single year, no multi-year production trend can be established yet; the score should be re-read as later vintages accumulate in OpenAlex.

Reproducibility55/100

Evolve's reproducibility axis averages 55 across 2 assessed papers (reproMean 55, reproPapers 2), each rated medium readiness. The verdicts converge on one shared limitation: PACT's gains rest on single-run evaluations, and Iris reports BrowseComp 82.2/88.6 without error bars. Reported numbers and theory look strong, but multi-seed replication evidence is absent from both papers.

Research summary

Yao Hu is a researcher at Xiaohongshu (Little Red Book), China, whose visible bibliographic record — resolved from OpenAlex — is recent and concentrated in 2026. The snapshot attributes 4 works to this identity (countsByYear 2026: 4; recent3yWorks: 4), and Evolve has reviewed two arXiv papers with Yao Hu in a middle author position on both. The profile splits evenly between two areas: machine learning (reinforcement learning for LLM post-training) and natural language processing / LLM agents (fieldDist 0.5/0.5). PACT: From Credit Assignment to Critic Alignment (2026) derives token-level credit as value martingale differences from three axioms, connecting actor-critic methods with PPO/GAE- and GRPO/RLOO-style setups; Evolve judges the theory sound and the gains strong while flagging single-run evaluations. Iris: Climbing to the Search Frontier (2026) builds open-source search agents from synthetic multi-hop data plus an SFT-RL climb, claiming BrowseComp scores of 82.2 and 88.6 — credible, though single-run and without error bars. OpenAlex topics span Information Retrieval and Search Behavior (3), Topic Modeling (2), Recommender Systems and Techniques (2), Web Data Mining and Analysis (2), and Natural Language Processing Techniques (1), indicating breadth across search- and recommendation-adjacent areas. Collaboration is mid-scale: 14 distinct coauthors and a recurring core of 12 frequent collaborators. Citation-based output and impact are not yet scored; h-index and i10-index both read 0 in the snapshot.

Collaboration pattern

The snapshot records 14 distinct coauthors and a recurring group of 12 frequent collaborators, including Zhaokai Luo, Mu Chuan, Jiayan Fu, Hang Xu, Yong Zhang, Dongyan Zhao, Ziyuan Liu, Hengqi Liu, Zichuan Wang, Yang Qin, Jiachen Liang, and Xu Chu. The computed collaboration score is 54, and both reviewed 2026 papers place Yao Hu in a middle author position rather than first or corresponding, consistent with team-based industry research.

Research keywords

token-level credit assignmentreinforcement learning for LLM post-trainingactor-critic methodsPPO/GAEGRPO/RLOOLLM search agentssynthetic multi-hop dataSFT-RL training pipelineBrowseComp evaluationinformation retrieval and search behavior

Highlights

PACT: three axioms uniquely fix token credit as value martingale differences
In this 2026 arXiv paper (middle authorship), Evolve finds that three axioms uniquely fix token-level credit as value martingale differences, with sound theory and strong PACT gains across PPO/GAE- and GRPO/RLOO-style setups; single-run evaluations limit confidence. Reproducibility readiness: medium.
Iris: claimed state-of-the-art open-source search agents on BrowseComp
In this 2026 arXiv paper (middle authorship), Evolve records SOTA open-source search agents built from synthetic multi-hop data plus an SFT-RL climb, with BrowseComp scores of 82.2 and 88.6 judged credible — though single-run and without error bars. Reproducibility readiness: medium.

Timeline

2026PACT: From Credit Assignment to Critic Alignment appears on arXiv; Evolve judges its axiomatic token-credit theory sound while flagging single-run evaluations.
2026Iris: Climbing to the Search Frontier appears on arXiv, claiming BrowseComp 82.2/88.6 for open-source search agents; results are judged credible but single-run without error bars.
2026OpenAlex attributes 4 works to this identity in 2026 (recent3yWorks: 4), marking the earliest visible year of the publication record.

Papers in the Evolve catalog

2026-09-22
PACT: From Credit Assignment to Critic Alignment
2026-09-03
Iris: Climbing to the Search Frontier

Frequent co-authors

Zhaokai Luo×2 Mu Chuan×2 Jiayan Fu×1 Hang Xu×1 Yong Zhang×1 Dongyan Zhao×1 Ziyuan Liu×1 Hengqi Liu×1 Zichuan Wang×1 Yang Qin×1 Jiachen Liang×1 Xu Chu×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

Bibliometrics come from OpenAlex only; output and impact dimensions are unscored because h-index and i10-index are 0 in the snapshot.
'Yao Hu' is a common name and no ORCID is provided; this profile is name-resolved from public bibliographic data and may merge or omit same-name authors.
Only 2 papers have been reviewed by Evolve; other indexed works are unassessed.
Both reviewed papers are 2026 arXiv preprints with no peer-review status recorded, and all evaluated results are single-run.